Global near real-time backscatter and InSAR products derived from Sentinel-1 geocoded bursts
Copernicus Programme’s Sentinel-1 SAR constellation images most of the land masses, with a revisit time of 6-24 days, in the Interferometric Wide (IW) swath Terrain Observation by Progressive Scanning (TOPS) mode. The S1 constellation has generated more than 10PB of Level-1 products since September 2014, and the size of this archive is expected to grow 3-4 fold over the next decade as more instruments are added to the constellation. Despite excellent global coverage and temporal sampling, continental and global scale processing campaigns with Sentinel-1 data have been hampered due to scalability of underlying SLC/GRD data access mechanisms (Appendix F of [1]). In this work, we present an efficient data access mechanism developed by Descartes Labs Inc. for Sentinel-1 SLC products and examples from the near real-time global backscatter and InSAR products that this capability has enabled.
Efficient SLC data access¤
We exploited the excellent burst synchronization properties of S1 TOPS acquisitions to build a global burst footprint map (See Figure 1) for all IW mode V-transmit data (~295K bursts), by analyzing all S1 annotation files from start of the mission till July 2020. We use this global burst map with every S1 SLC zip file released by ESA to label all burst footprints contained in it (new previously un-imaged footprints are added to global map), extract the SAR metadata for each burst and store it in json format for immediate access and identify the byte ranges of each individual burst in the zip files, using techniques developed in the neuroimaging and genomics communities [2], to enable random access to any burst from cloud storage buckets or public archives with minimal data transfer/ egress.
Note that our approach requires us to be able to access the complete zip file at least once in order to derive the exact byte ranges for the bursts. Once indexed, we are able to pull in metadata and imagery for any individual TOPS burst in ~3s from cloud storage or ~7s (without authentication time) from Alaska Satellite Facility. We have also developed tools to export these individual tools to formats compatible with open source tools like SNAP, GMTSAR and ISCE for use in various processing pipelines. We have been keeping up with ESA’s live stream of SLCs since mid September 2020 and have indexed all the historical SLCs over numerous country-scale AOIs with their metadata available for immediate use within large scale processing pipelines.
Geocoded bursts¤
Once we addressed the data access bottleneck, the next challenge to tackle was the use of custom Range-Doppler projections specific to every individual SLC image in SAR/InSAR processing workflows. SAR/InSAR workflows often pick one acquisition per imaging geometry as a reference and form a coregistered stack on this reference image’s grid. This imaging geometry is typically incompatible with the set of standard map projections like lat/lon or UTM or polar stereographic that are used in the geospatial community and GIS frameworks. To simplify our processing pipelines, we geocoded the SLC data with slant range phase corrections over an aligned UTM grid - 10 meter Northing and 2.5 meter Easting, similar to [3]. This then allows us to perform all interferometric and normalization operations in geographic space using simple raster operations. Note that our approach can easily incorporate a priori azimuth / slant range offset models during geocoding, if needed. Combined with our efficient data access mechanism, geocoding enables us to generate a coregistered stack for any burst footprint in the world in a few minutes. We have leveraged this newly developed capability to build two global data pipelines - one for SAR backscatter and one for InSAR coherence/ wrapped phase (see Figure 2). Both these pipelines have been keeping up with the live SLC data stream from ESA since late Jan 2021. We currently delete the geocoded bursts once we derive the necessary global products but can regenerate them very efficiently as needed by our higher resolution InSAR pipelines for infrastructure monitoring.
Global SAR backscatter product¤
Our Global SAR backscatter product is available at a posting of 10 meters on same UTM grid as Sentinel-2 and represents thermal noise corrected 𝜎0,E observations in dB. This product was derived using geocoded bursts, resulting in InSAR-grade coregistration of imagery acquired on different passes. We also provide three additional static layers which we generated using ESA’s SNAP tool Correction factor to transform 𝜎0,E in dB to 𝛾0,T (within 0.1 dB over most areas) Layover Shadow mask Local Incidence angle
We conducted numerous experiments with metadata from stacks of bursts to conclude that due to the small orbital tube characteristics for the Sentinel-1 mission, a static correction layer is more than sufficient for our applications where we desire Radiometrically Terrain Corrected backscatter data. Also, note that our global backscatter product is derived from the same source as our global InSAR product - geocoded SLC bursts, which results in better coregistration between these products. This may not be the case if these were generated using different approaches/ software -e.g., backscatter product with GRDs and the InSAR product with SLCs.
We will show examples of products from our live pipeline within our Viewer environment and present examples of our static layer based Terrain Correction, during the talk.
Global InSAR product¤
Our Global InSAR product is available at a posting of 20m on the same UTM grid as Sentinel-2 and is generated using a Gaussian Filter of wavelength 80m for multilooking and coherence estimation. Our coregistration approach is purely geometric for the global product and the processing setup ensures phase continuity at burst boundaries (see Figure 3). We currently generate all interferometric pairs with a temporal baseline of less than 25 days for live stream and have processed all pairs less than 13 days for historical data (before Nov 2020).
In addition to coherence, we preserved the wrapped phase at 20m which users can combine with the backscatter product to further average down spatially using simple band math (see Figure 3). Our burst based approach also lets users combine a consistent set of bursts to form their own standardized frames for mosaicking before phase unwrapping.
We will show examples of how co-seismic interferograms can be easily accessed and visualized from our global product in our Jupyter Lab-based workbench environment, during the talk.
Conclusions¤
Descartes Labs Inc has developed a very efficient random access mechanism for Sentinel-1 TOPS bursts, which enables it to generate global radar backscatter and InSAR products efficiently using geocoded bursts as an intermediate product. On a normal day of operations, these global layers are nominally updated well within 3-4 hours of SLCs being copied over from Copernicus programme’s scihub portal into our system. The developed data access mechanism that Descartes Labs Inc has developed eliminates SLC data access as the bottleneck in continental/ global scale processing / reprocessing campaigns. Pipeline throughput is now determined by the amount of compute resources and optimization of the analysis algorithms assigned to different pipelines rather than data access. For targeted infrastructure monitoring studies, we are able to generate coregistered, geocoded stacks of SLCs for any AOI in the world in a few minutes."